Loading...
1/* internal.h: mm/ internal definitions
2 *
3 * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11#ifndef __MM_INTERNAL_H
12#define __MM_INTERNAL_H
13
14#include <linux/mm.h>
15
16void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
17 unsigned long floor, unsigned long ceiling);
18
19static inline void set_page_count(struct page *page, int v)
20{
21 atomic_set(&page->_count, v);
22}
23
24/*
25 * Turn a non-refcounted page (->_count == 0) into refcounted with
26 * a count of one.
27 */
28static inline void set_page_refcounted(struct page *page)
29{
30 VM_BUG_ON(PageTail(page));
31 VM_BUG_ON(atomic_read(&page->_count));
32 set_page_count(page, 1);
33}
34
35static inline void __put_page(struct page *page)
36{
37 atomic_dec(&page->_count);
38}
39
40extern unsigned long highest_memmap_pfn;
41
42/*
43 * in mm/vmscan.c:
44 */
45extern int isolate_lru_page(struct page *page);
46extern void putback_lru_page(struct page *page);
47
48/*
49 * in mm/page_alloc.c
50 */
51extern void __free_pages_bootmem(struct page *page, unsigned int order);
52extern void prep_compound_page(struct page *page, unsigned long order);
53#ifdef CONFIG_MEMORY_FAILURE
54extern bool is_free_buddy_page(struct page *page);
55#endif
56
57
58/*
59 * function for dealing with page's order in buddy system.
60 * zone->lock is already acquired when we use these.
61 * So, we don't need atomic page->flags operations here.
62 */
63static inline unsigned long page_order(struct page *page)
64{
65 /* PageBuddy() must be checked by the caller */
66 return page_private(page);
67}
68
69/* mm/util.c */
70void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
71 struct vm_area_struct *prev, struct rb_node *rb_parent);
72
73#ifdef CONFIG_MMU
74extern long mlock_vma_pages_range(struct vm_area_struct *vma,
75 unsigned long start, unsigned long end);
76extern void munlock_vma_pages_range(struct vm_area_struct *vma,
77 unsigned long start, unsigned long end);
78static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
79{
80 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
81}
82
83/*
84 * Called only in fault path via page_evictable() for a new page
85 * to determine if it's being mapped into a LOCKED vma.
86 * If so, mark page as mlocked.
87 */
88static inline int is_mlocked_vma(struct vm_area_struct *vma, struct page *page)
89{
90 VM_BUG_ON(PageLRU(page));
91
92 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
93 return 0;
94
95 if (!TestSetPageMlocked(page)) {
96 inc_zone_page_state(page, NR_MLOCK);
97 count_vm_event(UNEVICTABLE_PGMLOCKED);
98 }
99 return 1;
100}
101
102/*
103 * must be called with vma's mmap_sem held for read or write, and page locked.
104 */
105extern void mlock_vma_page(struct page *page);
106extern void munlock_vma_page(struct page *page);
107
108/*
109 * Clear the page's PageMlocked(). This can be useful in a situation where
110 * we want to unconditionally remove a page from the pagecache -- e.g.,
111 * on truncation or freeing.
112 *
113 * It is legal to call this function for any page, mlocked or not.
114 * If called for a page that is still mapped by mlocked vmas, all we do
115 * is revert to lazy LRU behaviour -- semantics are not broken.
116 */
117extern void __clear_page_mlock(struct page *page);
118static inline void clear_page_mlock(struct page *page)
119{
120 if (unlikely(TestClearPageMlocked(page)))
121 __clear_page_mlock(page);
122}
123
124/*
125 * mlock_migrate_page - called only from migrate_page_copy() to
126 * migrate the Mlocked page flag; update statistics.
127 */
128static inline void mlock_migrate_page(struct page *newpage, struct page *page)
129{
130 if (TestClearPageMlocked(page)) {
131 unsigned long flags;
132
133 local_irq_save(flags);
134 __dec_zone_page_state(page, NR_MLOCK);
135 SetPageMlocked(newpage);
136 __inc_zone_page_state(newpage, NR_MLOCK);
137 local_irq_restore(flags);
138 }
139}
140
141#ifdef CONFIG_TRANSPARENT_HUGEPAGE
142extern unsigned long vma_address(struct page *page,
143 struct vm_area_struct *vma);
144#endif
145#else /* !CONFIG_MMU */
146static inline int is_mlocked_vma(struct vm_area_struct *v, struct page *p)
147{
148 return 0;
149}
150static inline void clear_page_mlock(struct page *page) { }
151static inline void mlock_vma_page(struct page *page) { }
152static inline void mlock_migrate_page(struct page *new, struct page *old) { }
153
154#endif /* !CONFIG_MMU */
155
156/*
157 * Return the mem_map entry representing the 'offset' subpage within
158 * the maximally aligned gigantic page 'base'. Handle any discontiguity
159 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
160 */
161static inline struct page *mem_map_offset(struct page *base, int offset)
162{
163 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
164 return pfn_to_page(page_to_pfn(base) + offset);
165 return base + offset;
166}
167
168/*
169 * Iterator over all subpages within the maximally aligned gigantic
170 * page 'base'. Handle any discontiguity in the mem_map.
171 */
172static inline struct page *mem_map_next(struct page *iter,
173 struct page *base, int offset)
174{
175 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
176 unsigned long pfn = page_to_pfn(base) + offset;
177 if (!pfn_valid(pfn))
178 return NULL;
179 return pfn_to_page(pfn);
180 }
181 return iter + 1;
182}
183
184/*
185 * FLATMEM and DISCONTIGMEM configurations use alloc_bootmem_node,
186 * so all functions starting at paging_init should be marked __init
187 * in those cases. SPARSEMEM, however, allows for memory hotplug,
188 * and alloc_bootmem_node is not used.
189 */
190#ifdef CONFIG_SPARSEMEM
191#define __paginginit __meminit
192#else
193#define __paginginit __init
194#endif
195
196/* Memory initialisation debug and verification */
197enum mminit_level {
198 MMINIT_WARNING,
199 MMINIT_VERIFY,
200 MMINIT_TRACE
201};
202
203#ifdef CONFIG_DEBUG_MEMORY_INIT
204
205extern int mminit_loglevel;
206
207#define mminit_dprintk(level, prefix, fmt, arg...) \
208do { \
209 if (level < mminit_loglevel) { \
210 printk(level <= MMINIT_WARNING ? KERN_WARNING : KERN_DEBUG); \
211 printk(KERN_CONT "mminit::" prefix " " fmt, ##arg); \
212 } \
213} while (0)
214
215extern void mminit_verify_pageflags_layout(void);
216extern void mminit_verify_page_links(struct page *page,
217 enum zone_type zone, unsigned long nid, unsigned long pfn);
218extern void mminit_verify_zonelist(void);
219
220#else
221
222static inline void mminit_dprintk(enum mminit_level level,
223 const char *prefix, const char *fmt, ...)
224{
225}
226
227static inline void mminit_verify_pageflags_layout(void)
228{
229}
230
231static inline void mminit_verify_page_links(struct page *page,
232 enum zone_type zone, unsigned long nid, unsigned long pfn)
233{
234}
235
236static inline void mminit_verify_zonelist(void)
237{
238}
239#endif /* CONFIG_DEBUG_MEMORY_INIT */
240
241/* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
242#if defined(CONFIG_SPARSEMEM)
243extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
244 unsigned long *end_pfn);
245#else
246static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
247 unsigned long *end_pfn)
248{
249}
250#endif /* CONFIG_SPARSEMEM */
251
252#define ZONE_RECLAIM_NOSCAN -2
253#define ZONE_RECLAIM_FULL -1
254#define ZONE_RECLAIM_SOME 0
255#define ZONE_RECLAIM_SUCCESS 1
256#endif
257
258extern int hwpoison_filter(struct page *p);
259
260extern u32 hwpoison_filter_dev_major;
261extern u32 hwpoison_filter_dev_minor;
262extern u64 hwpoison_filter_flags_mask;
263extern u64 hwpoison_filter_flags_value;
264extern u64 hwpoison_filter_memcg;
265extern u32 hwpoison_filter_enable;
1/* internal.h: mm/ internal definitions
2 *
3 * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11#ifndef __MM_INTERNAL_H
12#define __MM_INTERNAL_H
13
14#include <linux/fs.h>
15#include <linux/mm.h>
16
17void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
18 unsigned long floor, unsigned long ceiling);
19
20static inline void set_page_count(struct page *page, int v)
21{
22 atomic_set(&page->_count, v);
23}
24
25extern int __do_page_cache_readahead(struct address_space *mapping,
26 struct file *filp, pgoff_t offset, unsigned long nr_to_read,
27 unsigned long lookahead_size);
28
29/*
30 * Submit IO for the read-ahead request in file_ra_state.
31 */
32static inline unsigned long ra_submit(struct file_ra_state *ra,
33 struct address_space *mapping, struct file *filp)
34{
35 return __do_page_cache_readahead(mapping, filp,
36 ra->start, ra->size, ra->async_size);
37}
38
39/*
40 * Turn a non-refcounted page (->_count == 0) into refcounted with
41 * a count of one.
42 */
43static inline void set_page_refcounted(struct page *page)
44{
45 VM_BUG_ON_PAGE(PageTail(page), page);
46 VM_BUG_ON_PAGE(atomic_read(&page->_count), page);
47 set_page_count(page, 1);
48}
49
50static inline void __get_page_tail_foll(struct page *page,
51 bool get_page_head)
52{
53 /*
54 * If we're getting a tail page, the elevated page->_count is
55 * required only in the head page and we will elevate the head
56 * page->_count and tail page->_mapcount.
57 *
58 * We elevate page_tail->_mapcount for tail pages to force
59 * page_tail->_count to be zero at all times to avoid getting
60 * false positives from get_page_unless_zero() with
61 * speculative page access (like in
62 * page_cache_get_speculative()) on tail pages.
63 */
64 VM_BUG_ON_PAGE(atomic_read(&page->first_page->_count) <= 0, page);
65 if (get_page_head)
66 atomic_inc(&page->first_page->_count);
67 get_huge_page_tail(page);
68}
69
70/*
71 * This is meant to be called as the FOLL_GET operation of
72 * follow_page() and it must be called while holding the proper PT
73 * lock while the pte (or pmd_trans_huge) is still mapping the page.
74 */
75static inline void get_page_foll(struct page *page)
76{
77 if (unlikely(PageTail(page)))
78 /*
79 * This is safe only because
80 * __split_huge_page_refcount() can't run under
81 * get_page_foll() because we hold the proper PT lock.
82 */
83 __get_page_tail_foll(page, true);
84 else {
85 /*
86 * Getting a normal page or the head of a compound page
87 * requires to already have an elevated page->_count.
88 */
89 VM_BUG_ON_PAGE(atomic_read(&page->_count) <= 0, page);
90 atomic_inc(&page->_count);
91 }
92}
93
94extern unsigned long highest_memmap_pfn;
95
96/*
97 * in mm/vmscan.c:
98 */
99extern int isolate_lru_page(struct page *page);
100extern void putback_lru_page(struct page *page);
101extern bool zone_reclaimable(struct zone *zone);
102
103/*
104 * in mm/rmap.c:
105 */
106extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
107
108/*
109 * in mm/page_alloc.c
110 */
111extern void __free_pages_bootmem(struct page *page, unsigned int order);
112extern void prep_compound_page(struct page *page, unsigned long order);
113#ifdef CONFIG_MEMORY_FAILURE
114extern bool is_free_buddy_page(struct page *page);
115#endif
116extern int user_min_free_kbytes;
117
118#if defined CONFIG_COMPACTION || defined CONFIG_CMA
119
120/*
121 * in mm/compaction.c
122 */
123/*
124 * compact_control is used to track pages being migrated and the free pages
125 * they are being migrated to during memory compaction. The free_pfn starts
126 * at the end of a zone and migrate_pfn begins at the start. Movable pages
127 * are moved to the end of a zone during a compaction run and the run
128 * completes when free_pfn <= migrate_pfn
129 */
130struct compact_control {
131 struct list_head freepages; /* List of free pages to migrate to */
132 struct list_head migratepages; /* List of pages being migrated */
133 unsigned long nr_freepages; /* Number of isolated free pages */
134 unsigned long nr_migratepages; /* Number of pages to migrate */
135 unsigned long free_pfn; /* isolate_freepages search base */
136 unsigned long migrate_pfn; /* isolate_migratepages search base */
137 bool sync; /* Synchronous migration */
138 bool ignore_skip_hint; /* Scan blocks even if marked skip */
139 bool finished_update_free; /* True when the zone cached pfns are
140 * no longer being updated
141 */
142 bool finished_update_migrate;
143
144 int order; /* order a direct compactor needs */
145 int migratetype; /* MOVABLE, RECLAIMABLE etc */
146 struct zone *zone;
147 bool contended; /* True if a lock was contended */
148};
149
150unsigned long
151isolate_freepages_range(struct compact_control *cc,
152 unsigned long start_pfn, unsigned long end_pfn);
153unsigned long
154isolate_migratepages_range(struct zone *zone, struct compact_control *cc,
155 unsigned long low_pfn, unsigned long end_pfn, bool unevictable);
156
157#endif
158
159/*
160 * This function returns the order of a free page in the buddy system. In
161 * general, page_zone(page)->lock must be held by the caller to prevent the
162 * page from being allocated in parallel and returning garbage as the order.
163 * If a caller does not hold page_zone(page)->lock, it must guarantee that the
164 * page cannot be allocated or merged in parallel.
165 */
166static inline unsigned long page_order(struct page *page)
167{
168 /* PageBuddy() must be checked by the caller */
169 return page_private(page);
170}
171
172/* mm/util.c */
173void __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
174 struct vm_area_struct *prev, struct rb_node *rb_parent);
175
176#ifdef CONFIG_MMU
177extern long __mlock_vma_pages_range(struct vm_area_struct *vma,
178 unsigned long start, unsigned long end, int *nonblocking);
179extern void munlock_vma_pages_range(struct vm_area_struct *vma,
180 unsigned long start, unsigned long end);
181static inline void munlock_vma_pages_all(struct vm_area_struct *vma)
182{
183 munlock_vma_pages_range(vma, vma->vm_start, vma->vm_end);
184}
185
186/*
187 * Called only in fault path, to determine if a new page is being
188 * mapped into a LOCKED vma. If it is, mark page as mlocked.
189 */
190static inline int mlocked_vma_newpage(struct vm_area_struct *vma,
191 struct page *page)
192{
193 VM_BUG_ON_PAGE(PageLRU(page), page);
194
195 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
196 return 0;
197
198 if (!TestSetPageMlocked(page)) {
199 mod_zone_page_state(page_zone(page), NR_MLOCK,
200 hpage_nr_pages(page));
201 count_vm_event(UNEVICTABLE_PGMLOCKED);
202 }
203 return 1;
204}
205
206/*
207 * must be called with vma's mmap_sem held for read or write, and page locked.
208 */
209extern void mlock_vma_page(struct page *page);
210extern unsigned int munlock_vma_page(struct page *page);
211
212/*
213 * Clear the page's PageMlocked(). This can be useful in a situation where
214 * we want to unconditionally remove a page from the pagecache -- e.g.,
215 * on truncation or freeing.
216 *
217 * It is legal to call this function for any page, mlocked or not.
218 * If called for a page that is still mapped by mlocked vmas, all we do
219 * is revert to lazy LRU behaviour -- semantics are not broken.
220 */
221extern void clear_page_mlock(struct page *page);
222
223/*
224 * mlock_migrate_page - called only from migrate_page_copy() to
225 * migrate the Mlocked page flag; update statistics.
226 */
227static inline void mlock_migrate_page(struct page *newpage, struct page *page)
228{
229 if (TestClearPageMlocked(page)) {
230 unsigned long flags;
231 int nr_pages = hpage_nr_pages(page);
232
233 local_irq_save(flags);
234 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
235 SetPageMlocked(newpage);
236 __mod_zone_page_state(page_zone(newpage), NR_MLOCK, nr_pages);
237 local_irq_restore(flags);
238 }
239}
240
241extern pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
242
243#ifdef CONFIG_TRANSPARENT_HUGEPAGE
244extern unsigned long vma_address(struct page *page,
245 struct vm_area_struct *vma);
246#endif
247#else /* !CONFIG_MMU */
248static inline int mlocked_vma_newpage(struct vm_area_struct *v, struct page *p)
249{
250 return 0;
251}
252static inline void clear_page_mlock(struct page *page) { }
253static inline void mlock_vma_page(struct page *page) { }
254static inline void mlock_migrate_page(struct page *new, struct page *old) { }
255
256#endif /* !CONFIG_MMU */
257
258/*
259 * Return the mem_map entry representing the 'offset' subpage within
260 * the maximally aligned gigantic page 'base'. Handle any discontiguity
261 * in the mem_map at MAX_ORDER_NR_PAGES boundaries.
262 */
263static inline struct page *mem_map_offset(struct page *base, int offset)
264{
265 if (unlikely(offset >= MAX_ORDER_NR_PAGES))
266 return pfn_to_page(page_to_pfn(base) + offset);
267 return base + offset;
268}
269
270/*
271 * Iterator over all subpages within the maximally aligned gigantic
272 * page 'base'. Handle any discontiguity in the mem_map.
273 */
274static inline struct page *mem_map_next(struct page *iter,
275 struct page *base, int offset)
276{
277 if (unlikely((offset & (MAX_ORDER_NR_PAGES - 1)) == 0)) {
278 unsigned long pfn = page_to_pfn(base) + offset;
279 if (!pfn_valid(pfn))
280 return NULL;
281 return pfn_to_page(pfn);
282 }
283 return iter + 1;
284}
285
286/*
287 * FLATMEM and DISCONTIGMEM configurations use alloc_bootmem_node,
288 * so all functions starting at paging_init should be marked __init
289 * in those cases. SPARSEMEM, however, allows for memory hotplug,
290 * and alloc_bootmem_node is not used.
291 */
292#ifdef CONFIG_SPARSEMEM
293#define __paginginit __meminit
294#else
295#define __paginginit __init
296#endif
297
298/* Memory initialisation debug and verification */
299enum mminit_level {
300 MMINIT_WARNING,
301 MMINIT_VERIFY,
302 MMINIT_TRACE
303};
304
305#ifdef CONFIG_DEBUG_MEMORY_INIT
306
307extern int mminit_loglevel;
308
309#define mminit_dprintk(level, prefix, fmt, arg...) \
310do { \
311 if (level < mminit_loglevel) { \
312 printk(level <= MMINIT_WARNING ? KERN_WARNING : KERN_DEBUG); \
313 printk(KERN_CONT "mminit::" prefix " " fmt, ##arg); \
314 } \
315} while (0)
316
317extern void mminit_verify_pageflags_layout(void);
318extern void mminit_verify_page_links(struct page *page,
319 enum zone_type zone, unsigned long nid, unsigned long pfn);
320extern void mminit_verify_zonelist(void);
321
322#else
323
324static inline void mminit_dprintk(enum mminit_level level,
325 const char *prefix, const char *fmt, ...)
326{
327}
328
329static inline void mminit_verify_pageflags_layout(void)
330{
331}
332
333static inline void mminit_verify_page_links(struct page *page,
334 enum zone_type zone, unsigned long nid, unsigned long pfn)
335{
336}
337
338static inline void mminit_verify_zonelist(void)
339{
340}
341#endif /* CONFIG_DEBUG_MEMORY_INIT */
342
343/* mminit_validate_memmodel_limits is independent of CONFIG_DEBUG_MEMORY_INIT */
344#if defined(CONFIG_SPARSEMEM)
345extern void mminit_validate_memmodel_limits(unsigned long *start_pfn,
346 unsigned long *end_pfn);
347#else
348static inline void mminit_validate_memmodel_limits(unsigned long *start_pfn,
349 unsigned long *end_pfn)
350{
351}
352#endif /* CONFIG_SPARSEMEM */
353
354#define ZONE_RECLAIM_NOSCAN -2
355#define ZONE_RECLAIM_FULL -1
356#define ZONE_RECLAIM_SOME 0
357#define ZONE_RECLAIM_SUCCESS 1
358
359extern int hwpoison_filter(struct page *p);
360
361extern u32 hwpoison_filter_dev_major;
362extern u32 hwpoison_filter_dev_minor;
363extern u64 hwpoison_filter_flags_mask;
364extern u64 hwpoison_filter_flags_value;
365extern u64 hwpoison_filter_memcg;
366extern u32 hwpoison_filter_enable;
367
368extern unsigned long vm_mmap_pgoff(struct file *, unsigned long,
369 unsigned long, unsigned long,
370 unsigned long, unsigned long);
371
372extern void set_pageblock_order(void);
373unsigned long reclaim_clean_pages_from_list(struct zone *zone,
374 struct list_head *page_list);
375/* The ALLOC_WMARK bits are used as an index to zone->watermark */
376#define ALLOC_WMARK_MIN WMARK_MIN
377#define ALLOC_WMARK_LOW WMARK_LOW
378#define ALLOC_WMARK_HIGH WMARK_HIGH
379#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
380
381/* Mask to get the watermark bits */
382#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
383
384#define ALLOC_HARDER 0x10 /* try to alloc harder */
385#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
386#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
387#define ALLOC_CMA 0x80 /* allow allocations from CMA areas */
388#define ALLOC_FAIR 0x100 /* fair zone allocation */
389
390#endif /* __MM_INTERNAL_H */